How does the Swiss economy develop on a daily basis? Under the extraordinary circumstances caused by the Covid-19 pandemic, the economy is changing faster than usual but standard economic indicators cannot capture this. Indicators such as consumer sentiment index, GDP, or private consumption become available with a lag of up to several months.
As a solution, we provide a set of economic indicators for Switzerland based on Google search trends. These indicators are updated daily and provide policymakers and business leaders with timely information about the Swiss economy. Here you find more information on how to use our indicators and you can download the data.
The indicator for Perceived Economic Situation includes search terms that reflect people’s worries about the economy. For instance, people then google “economic crisis” (Wirtschaftskrise).
The category Watches and Jewellery includes stores and brands selling luxurious watches and jewellery goods and related, more general search terms for luxury consumer goods.
The category Mobility includes search terms related to ground transportation: For instance, checking the railway schedule or calling a taxi.
The category Travel Abroad includes search terms used to book flights and holidays.
The category Cultural Events includes search terms related to concerts, theatres, cinema and ticket providers for such events.
The category Gardening and Home Improvement includes stores selling materials for home improvement such as building materials, garden accessories and electrical supplies.
The category Clothing and Shoes includes clothing and shoe stores as well as a general search terms related to buying clothes and shoes. We found that people directly google for the brands they like. Note: the searchword “zalando” was not included because Zalando was not available in Switzerland before 2011.
The category Food Delivery includes search terms related to take away and ordering pizza.
Official economic statistics are usually released with a considerable lag. For example, the consumer sentiment index, GDP, or private consumption data become available with a lag of up to several months. The pace at which economic conditions are currently changing, however, requires more timely economic indicators in order to monitor and forecast economic activity. Economists have come up with different innovative ways to monitor the economy based on daily data, e.g. using daily electricity consumption.
To provide a set of timely economic indicators, we source Google Trends data. These indicators are available on a daily basis. The idea is that people’s searches on Google tell a lot about their perception of the state of the economy and demand for certain products and services.
When people become unemployed or fear they might lose their job, for example, they start searching for words like “arbeitslos” (unemployed). When people want to buy clothes, they likely type the name of fashion stores like “H&M” or “Zara”. Changes in these searches will then tell us something about unemployment or the demand for clothing.
These indicators provide insights about the abrupt economic changes we are currently experiencing. They will likely be among the first to capture changes in consumption once measures to fight the spread of Covid-19 are lifted step by step. They will thus help to monitor the recovery and possibly indicate lasting changes in consumption habits.
Our main indicator on the Perceived Economic Situation captures macroeconomic fluctuations. The other indicators focus on private consumption because i) of its overall importance for economic growth: private consumption makes up over 50% of Swiss GDP; ii) in this crisis, private consumption is hit hardest: due to closures of restaurants, stores, cinemas etc. private consumption is severely limited. This situation is very different to a financial crisis like the one that occurred in 2008/2009 and mainly hit the supply side of the Swiss economy.
To ensure that our combinations of keywords lead to meaningful results, we compare our indicators to existing measures, namely the consumer sentiment index, GDP, and components of private consumption expenditure whenever they provide a meaningful comparison. Our indicators broadly coincide with existing economic time series over the past 13 years, including the recession in 2009. However, note that our indicators do not attempt to replicate or replace any existing time series.
The figure below shows how our main indicator on the perceived economic situation compares to quarterly GDP growth in Switzerland and the consumer confidence index. For comparison, we have aggregated our indicator to quarters. In this figure, the last observation of our indicators is based on Google searches from January to March 2020, the first quarter of the year. This explains why the drop looks less dramatic than when looking at daily data.
For some indicators, such as Cultural Events – which includes demand for restaurants, cinemas, concerts etc. – or Gardening and Home Improvement (aka do-it-yourself articles), there is no preexisting, well-established indicator to compare our search-based indicators with. Nevertheless these indicators appear meaningful, as we are observing substantial changes in searches for these goods and services compared to normal times. These indicators will therefore provide valuable information on whether the economy is going back to normal as lock-down measures are lifted. In the beginning, some people, especially people in risk groups, may still be be reluctant to go, for example, to restaurants or stroll around shopping malls because they are afraid of contracting the virus.
The relationship between these search-based indicators and economic activity seems apparent. We believe that these indicators will provide useful during the whole economic roller-coaster ride that lies ahead of us, during the current downturn as well as the upswing(s) when things go slowly back to (a new?) normal.
Note however, that the relationship between Google searches for specific keywords and real economic activity may change over time. First, the popularity of specific internet platforms and online stores can change, and in that case Google searches as well as actual consumption behavior will change too.
Second, this crisis is very different to previous ones as private demand drops largely because of mandated closures of restaurants, cinemas and stores to fight the spread of Covid-19.
Third, the crisis itself might lead to permanent changes in search behavior. For example, containment measures may lead to a permanent increase of the proportion of the population doing grocery shopping online. While there are currently two major supermarkets offering online shopping, people might start to search more and more for other shops offering this service.
Die erste Jahreshälfte 2020 darf schon jetzt als historisch betrachtet werden. Selten zuvor sind Volkswirtschaften in derart kurzer Zeit so sehr eingebrochen wie jetzt. Infolge von COVID-19 und den diesbezüglichen staatlichen Massnahmen steigen praktisch überall die Arbeitslosenzahlen rapide an. Die wirtschaftliche Leistung gemessen am Bruttoinlandsprodukt (BIP) fällt stärker als wir es seit Jahrzehnten gesehen haben. Nach wie vor herrscht jedoch grosse Unsicherheit darüber, wie dramatisch der wirtschaftliche Einbruch wird und wie rasch sich Volkswirtschaften von der Krise erholen werden.
In der Schweiz ist ein Rückgang des BIP um rund sechs Prozent zu erwarten (Seco, April 2020). Allerdings trifft dies nicht alle Sektoren der Wirtschaft gleichmässig. Restaurants oder auch der Flughafen Zürich sind ungleich härter betroffen als beispielsweise Supermärkte oder die Pharmabranche. Während die Extrema recht offensichtlich sind, bleiben jedoch viele Fragen offen.
Wie schnell steigt beispielsweise die Nachfrage nach langlebigen Konsumgütern wieder? In Rezessionen sparen Konsumentinnen und Konsumenten insbesondere bei Kleidung, Möbeln oder Autos. Ausgaben in diesen Bereichen lassen sich leichter aufschieben, weshalb diese Branchen sehr pro-zyklisch sind: In der Rezession leiden sie besonders – und mit ihnen die Werbebranche und in nächster Instanz Zeitungen, welche weniger Werbeanzeigen verkaufen. Somit ist die Nachfrage nach Kleidung oder Autos über Zweit- und Drittrundeneffekte auch für viele andere Industrien wichtig.
Sowohl für die Politik als auch für Unternehmen stellt die hohe Unsicherheit ein Problem da. Wie viel staatliche Unterstützung braucht es und wo ist sie am nötigsten? Ab wann können sich Firmen wieder auf steigende Nachfrage und damit Umsätze einstellen?
Daten können uns helfen, solche Fragen besser zu beantworten. Allerdings müssen diese vielfach erst aufbereitet und graphisch veranschaulicht werden. Die von uns neu geschaffene Plattform trendEcon.org setzt sich genau dies zum Ziel. Jeden Tag suchen Schweizerinnen und Schweizer auf Google nach unzähligen Begriffen. Via Google Trends lässt sich für jeden einsehen, welche Begriffe derzeit sehr häufig gesucht werden. Illustrativ ist ein Blick auf die Begriffe «WC Papier» sowie «Rezession»:
Es zeigt sich hier anschaulich, wie ab der zweiten Märzwoche die Suchanfragen nach «Rezession» stiegen und viele Menschen Mitte März aufgrund einer kurzfristigen Knappheit an WC-Papier danach im Internet gesucht haben. Die Sorge um das WC-Papier ist rasch verflogen, aber das Thema Rezession ist nach wie vor sehr präsent, wenn auch auf einem tieferen Niveau.
Diese Beispiele verdeutlichen auch einige Probleme: Welche Suchbegriffe sind relevant und auf welche Weise können mehrere Begriffe kombiniert verwendet werden? Wie lassen sich die Daten sauber vergleichbar darstellen: was ist der sinnvolle Vergleichsmassstab?Diese Aspekte versuchen wir mit trendEcon zu adressieren, indem wir aus einer Vielzahl an möglichen Suchbegriffen die geeignetsten auswählen und die Daten ab Januar 2006 aufbereiten. Die Datenaufbereitung beinhaltet unter anderem eine saisonale Bereinigung ebenso wie die Berücksichtigung von Feiertagen.
Im Ergebnis zeigt trendEcon derzeit neben einem Krisenindex auch für mehrere Branchen die zeitliche Entwicklung auf. Unser Hauptindikator «Perceived Economic Situation» basiert auf den Google Suchanfragen in der Schweiz nach den Stichworten «Wirtschaftskrise», «Kurzarbeit», «arbeitslos» und «Insolvenz». Die darauf basierenden Daten liefern ein Bild davon ab, wie sehr sich die Menschen in der Schweiz um die wirtschaftliche Entwicklung sorgen. Je häufiger nach den vier Begriffen gesucht wird, desto schlechter die Einschätzung der aktuellen Situation. Graphisch zeigt sich deutlich der Einbruch ab Ende Februar und der Tiefpunkt kurz nach den Beschlüssen des Bundesrats vom 16. März 2020:
The indicator for Perceived Economic Situation includes search terms that reflect people’s worries about the economy. For instance, people then google “economic crisis” (Wirtschaftskrise).
Auffällig ist zudem, dass sich die Stimmung bereits wieder deutlich aufgehellt hat – wobei der Indikator nach wie vor unter dem langfristigen Niveau ist. Als sehr geeigneter Vergleich kann man die Daten aus dem Jahr 2019 heranziehen und erkennt, wie markant der Unterschied ist:
Neben der gesamtwirtschaftlichen Betrachtung bietet trendEcon auch eine Reihe von sektorspezifischen Indikatoren an. Dabei zeigt sich – als Bestätigung für die Qualität der Daten – der starke und anhaltende Einbruch der Reisebranche. Suchanfragen nach «Städtetrip», «Flug buchen» oder «günstige Flüge» sind nach wie vor viel geringer als zu Vor-Corona Zeiten:
The category Travel Abroad includes search terms used to book flights and holidays.
Einen ähnlichen und anhalten Einbruch sehen wir bei kulturellen Veranstaltungen. Im Gegensatz dazu ist die Nachfrage nach Essenslieferungen oder auch Baumärkten auf einem (positiven) Rekordstand.
Zusammengefasst lässt sich sagen, dass trendEcon eine Plattform für tagesaktuelle Informationen zum Zustand der schweizerischen Wirtschaft bietet. Wir werden in den kommenden Wochen weitere Suchbegriffe und (Sub-) Indikatoren hinzufügen, um das Angebot an Informationen stetig zu vergrössern. Ausserdem testen wir laufend die Nützlichkeit unserer Daten für die Prognose von volkswirtschaftlich relevanten Grössen. Sämtliche Daten sind frei zum Download verfügbar und der Code wird bald auf GitHub veröffentlicht.
May 25, 2020
The first half of 2020 can already be regarded as historic. Seldom before have economies shrunkas fast in such a short time. As a result of COVID-19 and the related government measures, unemployment figures are rising rapidly practically everywhere. Economic performance measured by gross domestic product (GDP) is falling more sharply than we have seen in decades. However, there is still vast uncertainty about how dramatic the economic slump will be and how quickly economies will recover from the crisis.
In Switzerland, annual GDP is expected to fall by around six percent (Seco, April 2020). However, this will not affect all sectors of the economy equally. Restaurants and airports are himuch harder hit than supermarkets or the pharmaceutical industry, for example. While the extremes are quite obvious, many questions remain open.
How quickly, for example, will the demand for durable goods rise again? In recessions, consumers save money, reducing spending especially on clothing, furniture, or cars. Spending in these areas is easier to postpone, which is why these sectors are very pro-cyclical: In a recession they suffer particularly - and with them the advertising industry and, in the next instance, newspapers, which sell fewer advertisements. Consequently, the demand for clothing or cars is also important for many other industries via second- and third-round effects.
The high level of uncertainty poses a problem for both politicians and companies. How much government support is needed and where is it most needed? When will companies again face increasing demand and thus turnover?
Data can help us to improvethe answersto such questions. However, in many case, data first have to be prepared and graphically illustrated. The new platform trendEcon.org created by us aims at exactly this. Every day, Swiss people search Google for countless keywords. Via Google Trends, anyone can see which keywords are currently being searched for very frequently. A look at the terms “WC Papier” (toilet paper) and “Rezession” (recession) is illustrative:
It can be clearly seen here how from the second week of March onwards the number of search requests for “recession” increased. We also observe the large number of people searching for toilet paper on the Internet in mid-March due to a short-term shortage. The concern about toilet paper quickly disappeared, but the topic of recession is still very present.
While such examples are very illustrative, they also highlight some of the problems: Which search terms are actually relevant and how can several terms be combined? How can the data be presented in a clean and comparable way? What are useful benchmarks? We at trendEcon try to address these aspects by selecting the most suitable search terms from a multitude of possible search terms and preparing the data from January 2006 onwards. The latter includes among others a seasonal adjustment as well as the consideration of public holidays.
As a result, trendEcon currently shows not only a crisis index but also the development over time for several sectors. Our main indicator “Perceived Economic Situation” is based on the keywords “economic crisis”, “short-time work”, “unemployed” and “insolvency”. The data based on this provides a picture of the extent to which people in Switzerland are concerned about economic development. The more frequently the four terms are searched for, the worse the assessment of the current situation. The slump from the end of February onwards and the low point shortly after the Federal Council’s decisions of March 16, 2020 are clearly visible in the graph:
The indicator for Perceived Economic Situation includes search terms that reflect people’s worries about the economy. For instance, people then google “economic crisis” (Wirtschaftskrise).
It is also noticeable that sentiment has already brightened considerably again - although the indicator is still below the long-term level. As comparison, one can use the data from 2019 to see the striking differ-ence:
In addition to the overall economic analysis, trendEcon also offers a range of sector-specific indicators. The strong and continuing slump in the travel industry is a confirmation of the quality of the data. Search queries for“city trip”, “book flight” or “cheap flights” (note: we use the German words) are still rare compared to the pre-Corona period:
The category Travel Abroad includes search terms used to book flights and holidays.
We see a similar and continuing slump in cultural events. In contrast, demand for food deliveries or DIY stores is at a (positive) record level.
In summary, trendEcon offers a platform for daily updated information on the state of the Swiss economy. In the coming weeks, we will be adding further search terms and (sub-) indicators in order to constantly increase the amount of information available. In addition, we are constantly testing the usefulness of our data for forecasting economically relevant variables. All data are freely available for Download and the code will be publicly available soon.
If you query Google Trends for a search term, e.g., insolvenz, the result is based on a subsample of all search results.
For a small country like Switzerland, these results may differ quite dramatically from sample to sample.
In order to alleviate the problem, we usually draw 12 or more samples for each series (keyword and frequency). We force Google to re-sample by choosing a different time window.
Google search results are available on a daily, weekly or monthly frequency. As with individual samples, querying Google at a different frequency will lead to substantially different results.
Our goal is to produce long daily time series, ideally from 2006. This is mostly because we want to use the financial crisis of 2008 and 2009 as a benchmark for the Covid-19 crisis. However, Google does not provide daily or weekly data for such a long time period. We circumvent the problem by applying a moving window of daily and weekly queries over the whole time period.
Since the daily, weekly and monthly series are still very different, we want to combine them into a single daily series.
Our next steps are based on the following assumptions:
Monthly data catches the long term trend in search activity in the most accurate way.
Weekly data is best to analyze the searches over the medium term, i.e., over a few months.
Daily data is best to analyze short term behavior over a few days and weeks.
With this in mind, we apply the following methodology. In a first step, we “bend” the daily series to the weekly values, by applying a variant of the Chow-Lin (1971) method. This preserves the movement of the daily series and ensures that weekly averages are identical to the original weekly series.
In a second step, we bend the adjusted weekly series from the first step to the monthly values, using the same method again. This produces a series that maintains most of the movement of the daily and the weekly series, but has the same monthly averages as the monthly series. This series are used for further processing described in the subsequent sections.
Daily series have multiple seasonalities:
There is weekly seasonality : Search activities for business related activities may lower during the weekend.
There is monthly seasonality : If salaries are paid by the end of the month, some kind of shopping may be more likely to occur afterwards.
There is yearly seasonality : Flip flop is an unpopular search term in Winter, while Christmas is rarely looked up in Summer.
Finally, there are irregular holidays that occur at a different dates each year, such as Easter.
We use the Prophet procedure for estimating an additive model where non-linear trends are fit with yearly and weekly seasonality and the holiday effects. The procedure is fully automated and easy to use, but has a few drawbacks. It does not model monthly seasonality and all the seasonal effects are assumed to be constant over time.
In a final step, we use principal component analysis (PCA) to extract the common signal from a group of seasonally adjusted, daily time series. The idea of PCA is to identify several principal components that summarize most of the variance in the data. We then use the first principal component as our index. We manually checked the factor loadings and scores and adjusted the variable selection if needed.
The data can be accessed here. Our indicator data is public and freely available under license Attribution 4.0 International CC BY 4.0. This means that you can remix, adapt, and build upon this work non-commercially, as long as you credit us and license your new creations under the identical terms.
This project was born at the end of March 2020 as economists were in dire need to obtain timely data to estimate the impact of the Covid-19 pandemic on the Swiss economy.
It started as a collaboration between economists working at the data consulting company cynkra, KOF Swiss Economic Institute, the economic forecast division of the State Secretariat for Economic Affairs (SECO), the Swiss Federation of Trade Unions (SGB) and the University of St. Gallen.
We will continue to work on this project along the way and possibly add further indicators.
On Google Trends, you will find series based on a random sample of Google searches conducted in Switzerland. Depending on the time frame you choose, Google returns daily, weekly, or monthly data. We usually downloaded the data 12 times or more times to construct long-run data with a daily frequency. We further adjust the data for seasonal patterns in searches. It is not surprising that searches for gardening are higher in spring. To make meaningful comparisons over time, these seasonal patterns are removed. Finally, our indicators are not based on one single keyword but several keywords are aggregated into one indicator. See Method for further details.
We carefully selected keywords based on expected consumer search behavior. For each indicator, we list the keywords next to the figure in the mini-tab Info.
We try to stick to those keywords that are most likely to reflect the intention to consume certain types of goods and services rather than just having an interest in learning more about them.
You’ll probably still find that not every keyword that one might think of went into our indicator. This is due to the following reasons: First, to ensure privacy, Google Trends does not return search trends based on a small number of searches. Some keywords of interest are therefore censored and return a lot of zeros. Second, other keywords do not show any significant changes over time that coincide with economic activity or changes in demand. These keywords contain no relevant information to go into our indicators. Finally, more is not always better. We use those keywords that correlate strongest with real economic activity.
German is the main language of 63% of the population in Switzerland. Search trends on Google are likely to be representative for the whole economy. Indeed, this is what we find if we compare our indicators to the existing target indicators like consumer sentiment. Due to Google’s privacy restrictions, searches in French and Italian within Switzerland are more often censored than German ones. If keywords are censored, Google replaces observations with a zero and we cannot use the keyword. This is another reason why we opt for using names of stores or brands: these do not depend on the language preferred by the Google user.
We put a lot of effort into constructing daily, seasonally adjusted indicators that correlate with existing economic indicators. They can be used to say something about the evolution of demand for certain consumption categories. However, they remain indicators. We do not measure true economic activity as in number of T-shirts sold on a specific day or money spent on gardening equipment. Also note that this is an experiment in real time. If the correlation between search behavior and consumer demand changes, our indicators may become better or get worse in tracking consumer demand.
The indicators are based on search volumes for chosen keywords. However, the number of searches is unknown, as Google does not share this information to protect users’ privacy. Depending on the indicator, you can interpret the indicators as relative changes over time in interest, concerns, or demand for certain topics, goods, or services. We list the keywords that went into the indicator to facilitate interpretation in the mini-tab Info next to the figure representing each indicator.
The more technical and quantitative interpretation is as follows: all series are normalized such that the long-term equals zero and the standard deviation is one. E.g., an index value of 2 means that search volume for this item is two standard deviations above the average.
The indicators cannot quantify changes in GDP or total consumer demand. Given how the indicators are constructed, it is also impossible to infer growth rates. Google does not share the total number of searches to protect users’ privacy makes it impossible to say something about absolute changes in search volumes.
The data can be accessed here. Our indicator data is public and freely available under license Attribution 4.0 International CC BY 4.0. This means that you can remix, adapt, and build upon this work non-commercially, as long as you credit us and license your new creations under the identical terms.